Combined Driver Distraction and Intention Algorithm for Maneuver Prediction and Collision Avoidance

Abstract

Driver intention detection holds high potential for adaptive driver assistance systems and automated driving functions. To develop a combined driver distraction and intention model as well as an intention detection algorithm a real driving study with 45 subjects performing 1260 braking and 1890 evasion maneuvers was conducted and analyzed. The driver‘s distraction level and hand position are varied to analyze their influence on driver intention. With a probabilistic approach, a sensitivity analysis of indicators for detecting driver intention was developed. The accelerator pedal and the longitudinal and lateral accelerations reveal to be most sensitive for evasion, while the longitudinal acceleration, the brake pressure and the accelerator pedal are most sensitive for braking. By using this sensitivities for algorithm design and combining them with information about whether drivers have recognized the object and their distraction level, evasion maneuvers can be detected correctly at least three seconds prior to passing the object in 91 % of all cases, braking maneuvers in 87 % of all cases. The driver‘s distraction level turned out to be relevant for intention recognition, as 87 % of drivers reduce their distraction at least three seconds prior to passing the object. We conclude that drivers cannot have a relevant intention and be highly distracted at the same time. Driver distraction detection hence contributes to the driver intention recognition. A three seconds prediction frame allow effective risk mitigation by warning and automated interventions.